Nate Silver - Trump's jobs data denialism won't fool anyone [View all]

If you’d asked me 20 years ago, when I made a living playing online poker and running projections for Baseball Prospectus, I would have told you that, of course, I’m a Data Guy. Treat the numbers with caution, because it’s easy to build bad models or otherwise screw up in umpteen ways when working with complex statistical data under deadline pressure. But at the end of the day, we all have to make decisions based on incomplete information. Businesses particularly face this problem: new hires or capital investments typically have a long time horizon. If you’re not going to make these choices based on your best estimate of the situation, given the relevant uncertainties — well, what exactly are you going to base them on?

These days, I’d say my relationship with the Data Guy label has become more fraught. In one of my particular subfields, election forecasting, there’s been a lot of bad work, which probably did more to misinform the conversation than contribute to it. And more broadly, I do buy to some extent that “science has been politicized”. Sometimes I've seen the admonition “just trust the numbers” used as a blunt insturment to suppress legitimate dissent. Data is collected by and interpreted by humans, and human error and bias play a role at every step of the process.

The important thing is to be right, and probably 90 percent of the time, going through the rigor of kicking the tires on the data2 and then explicitly modeling a situation helps you get there. But I’ll give a little more weight these days to the mesoscale rather than the micro. Having a good bullshit detector — heuristics honed by life experience — comes in handy that remaining 10 percent of the time.

When it comes to economic data produced by the United States government, however, people play that 10 percent “get out of jail free” card far too often. And usually, they’re the ones producing the bullshit, often to spin away politically inconvenient data.

https://www.natesilver.net/p/trumps-jobs-data-denialism-wont-fool